You know the 80/20 problem? The last 20% percent of the work require 80% of the effort? Well, with self-driving cars it's the reverse. Most start ups spend the first year getting a team of car hackers to hook up a bunch of ROS machines in the trunk to the car's controls and rigging up a bunch of sensors. And then once they're done, they have this one task left, writing the actual algorithms. And they realize that the best they can do is a car that makes laps around their parking lot (see above article), and even that takes them another year.
The 80/20 thing is not that... It's 80% of the outcomes are from 20% of the causes. Originally seen from 20% of the population holds 80% of the land in Italy.
In software you could say 80% of the bug reports come from 20% of the bugs.
There certainly is a right way to apply it. No where does that rule say the last 20 percent of reward takes 80 percent of the effort. That's grossly mischaracterizing the rule.
You know those five self-driving "levels"? Hands off, eyes off, attention off, mind off, or something like that? I think we need a sixth, "mind off, but this time you're driving in India".
Every time I go on the road, I am reminded that what works in the US has no way of working where I live, unless the car can also automatically honk its horn and cut people off. Otherwise, in high-traffic conditions it would just stay put behind the guy who double-parked in its lane for two hours, politely never cutting anyone off.
Yeah you don't even need to go to that extreme. I'm yet to see a self driving car handle the extremely common situation of meeting an oncoming car on a road that is effectively single lane because of cars parked on the side. Or just, and road layout more complicated than US-style grids.
This is pretty neat, perhaps the author should integrate with the comma.ai platform, which is open source (except for the vision part, which this post implements), has a vehicle dynamics model and hardware integration with several recent Honda and Toyota cars.
I wish them well, but they're doing hobby robotics.
They claim to be at the level of Waymo/Tesla/Uber/Cruise, etc, but a quick glance at their github shows sub-DARPA Grand Challenge levels of technical development.
You are likely to provide more of the actual things needed to make a self-driving car work than any car company is, though. No car company will give the engineers the time, resources, or control over the project to make a self-driving car that is trustworthy. But an individual might at least provide themselves a quiet environment, the tools they need, the time they need, and never be forced to rush or forge ahead with uncomfortable reservations simply because an MBA wants to meet his metric projections and knows that no court in the land would ever levy criminal charges no matter what happens.
